Today's general-purpose graphics processing units (GPGPUs) have dedicated walkers for both media and compute workloads. However, in conventional techniques, for any given workloads, the walking patterns are fixed that often lead to inefficient thread launching which, in turn, creates inefficient load balancing. Further, conventional techniques rely on hardcoded walking patterns and thread launching, which is not ideal for various applications, such as those having varying patterns.